Automated assessment of bilateral breast volume asymmetry as a breast cancer biomarker during mammographic screening

Transcription

1 Automated assessment of bilateral breast volume asymmetry as a breast cancer biomarker during mammographic screening Alex C. Williams b, Austin Hitt b, Sophie Voisin a, Georgia Tourassi* 1a a Biomedical Science and Engineering Center, Oak Ridge National Laboratory, Oak Ridge, TN USA 37831; b Department of Computer Science, Middle Tennessee State University, Murfressboro, TN USA ABSTRACT The biological concept of bilateral symmetry as a marker of developmental stability and good health is well established. Although most individuals deviate slightly from perfect symmetry, humans are essentially considered bilaterally symmetrical. Consequently, increased fluctuating asymmetry of paired structures could be an indicator of disease. There are several published studies linking bilateral breast size asymmetry with increased breast cancer risk. These studies were based on radiologists manual measurements of breast size from mammographic images. We aim to develop a computerized technique to assess fluctuating breast volume asymmetry in screening mammograms and investigate whether it correlates with the presence of breast cancer. Using a large database of screening mammograms with known ground truth we applied automated breast region segmentation and automated breast size measurements in CC and MLO views using three well established methods. All three methods confirmed that indeed patients with breast cancer have statistically significantly higher fluctuating asymmetry of their breast volumes. However, statistically significant difference between patients with cancer and benign lesions was observed only for the MLO views. The study suggests that automated assessment of global bilateral asymmetry could serve as a breast cancer risk biomarker for women undergoing mammographic screening. Such biomarker could be used to alert radiologists or computer-assisted detection (CAD) systems to exercise increased vigilance if higher than normal cancer risk is suspected. Keywords: mammography, breast cancer, diagnosis, fluctuating asymmetry, imaging biomarker 1. INTRODUCTION Bilateral symmetry of paired structures (i.e., hands, ears, or breasts) is well established as an important marker of developmental stability and good health in humans [1]. Although small deviations from perfect symmetry is normal, several studies have linked high fluctuating asymmetry to a variety of genetic and mental disorders. Many of these studies are reviewed in Ref [1]. Fluctuating asymmetry (FA) is the scientific term that describes the random deviations from perfect bilateral symmetry. Sexually selected traits are more susceptible to high FA [2-4]. Thus, any developmental or health disturbance could be potentially revealed in higher asymmetries of sexually selected traits. Visually apparent bilateral differences in breast shape and size are relatively common in women although we have not found any reported statistics on their prevalence. Typically, these differences manifestate during puberty when estrogen hormone stimulates breast growth. In some women, the breasts simply grow differentially with respect to starting and stopping. In these young women, breast size asymmetry is considered a sign of developmental instability. 1 tourassig@ornl.gov Notice: This manuscript has been authored by UT-Battelle, LLC, under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. Medical Imaging 2013: Computer-Aided Diagnosis, edited by Carol L. Novak, Stephen Aylward, Proc. of SPIE Vol. 8670, 86701A 2013 SPIE CCC code: /13/$18 doi: / Proc. of SPIE Vol A-1

2 Breast size asymmetry as an indicator for breast cancer susceptibility has been explored before. In 1997, bilateral breast asymmetry was linked to the presence of breast cancer [5]. The study showed significantly higher fluctuating asymmetry of the breasts in breast cancer patients than in an age-matched control group. Since the study found no correlation between tumor size and amount of breast size asymmetry, the authors argued that their finding could be more than merely the effect of cancer. It could be a predictor of future breast cancer risk. A 2006 study by the same authors showed that indeed breast size asymmetry is linked to predisposition to breast cancer [6]. The study was based on 504 age-matched asymptomatic women who had normal mammograms. One half of those women later developed breast cancer. The study showed larger breast volume bilateral asymmetry in those women than the control group (i.e., women with normal mammograms during the study period). For the post-menopausal subgroup, breast volume asymmetry assessed in the craniocaudal (CC) view was found to be a significant predictor of breast cancer. Both studies were based on breast size measurements made manually on film mammograms by expert radiologists using a ruler. A study, published in 2000, linked ipsilateral decrease in breast size to infiltrating lobular carcinoma (ILC) [7]. A similar finding was reported for clinically advanced ILC cases in 1991 [8]. The study showed that a similar trend existed for ILC cases that were not advanced cancers. Although the breast size decrease was mammographically measurable, it was not physically apparent. These studies provide further evidence that the presence of breast cancer (or at least some types of breast cancer) causes changes in breast size that are measurable in mammography. Ipsilateral breast size changes would translate to higher bilateral breast asymmetry. The above studies were based on radiologists ruler-based measurements of breast size from mammograms. The purpose of our study was to apply an automated way of measuring fluctuating breast volume asymmetry from mammograms and investigate whether the automated measurements correlate with the presence of breast cancer using a large dataset of screening mammograms with known ground truth. The issue of bilateral breast differences has been investigated only in the context of regional matching for localized detection of breast abnormalities [e.g., 9-14]. Typically, a mammographic region of interest is identified and then compared to its matched location in the bilateral breast to determine if it is different enough to qualify as a detected abnormality. Reference 15 provides a recent survey of studies of algorithms for alignment of the left and right breasts as well as methods for comparing the morphology, texture, and directionality of the left and right breast parenchyma. Unlike those studies, we aim to explore automated assessment of global bilateral breast asymmetry in size and volume as a diagnostic indicator of breast cancer. 2.1 Breast Size Measurements 2. MATERIALS AND METHODS Breast region segmentation is a critical step for automated assessment of breast size. Numerous algorithms have been proposed for the task, utilizing thresholding, gradients, polynomial modeling, active contours, and many others [16]. For this study we implemented the algorithm proposed by Mirzaalian et al [17] for segmenting the breast region in both the craniocaudal (CC) and medio-lateral oblique (MLO) views of a screening mammogram. The algorithm includes lowlevel preprocessing such as histogram equalization, convolution with a low-pass mask, thresholding, connectedcomponent analysis, as well as a non-linear diffusion algorithm for segmenting the pectoral muscle in the MLO views. The software was developed in opencv. The height and width of each segmented breast were determined automatically in each view. Figure 1 illustrates these breast size measurements in each view. Then, the volume of each breast was calculated according to well-established, published methods, which have been shown to be highly reliable. Namely, the Katariya method was used in CC views [18], the Hoe method for MLO views [19], and the Fung method was applied to assess breast volume size by fusing breast size measurements from both the CC and MLO views [20]. Proc. of SPIE Vol A-2

3 CC NIL() w Figure 1: Illustration of breast size measurements in CC and MLO views Katariya Method [18]: This method focuses on the craniocaudal (CC) view and essentially aims to automate the manual measurement process followed in [5]. Specifically, breast height (h) is measured as the maximum perpendicular distance from the distal edge of the breast outline to the proximal edge of the film. Breast width (w) is measured as the maximum breast width across the proximal edge of the film. Finally, the breast volume is calculated assuming a circular cone shape for the breast: CC: Volume = " # w 2 CC # h CC 12 (1) Hoe Method [19]: This method focuses on the mediolateral oblique (MLO) view and essentially automates the manual measurement process followed in [20]. Specifically, breast height (h) is measured as the maximum perpendicular distance from the distal edge of the breast outline to the pectoral muscle. Breast width is measured as the superolateralto-inferomedial perpendicular distance at full (w) and half height (z). MLO: Volume = h MLO " z MLO (2) Fung Method [21]: Recently Fung et al proposed a modified version of the Katariya algorithm, treating the breast as an elliptical (instead of circular) cone on both CC and MLO views (instead of CC only). Using the height measurement from the MLO view aimed to eliminate a potential error introduced in the CC view which may miss the base of the breast and underestimate the breast volume. CC+MLO: Volume = " # w 2 CC # h MLO 12 (3) The above methods do not take into account the breast compression factor during the image acquisition process. Although volume calculations that include the compression factor are slightly more accurate, studies have shown that the formulas given above correlate highly with true breast measurements made on mastectomy specimens [22]. 2.2 Fluctuating Asymmetry The above size parameters are calculated separately for each breast and from each mammographic view. Since the study focuses on bilateral breast differences, we calculate the absolute volume difference between the left (L) and right (R) breasts. The absolute difference can be utilized as a measurement of fluctuating asymmetry (FA). It has been recommended to normalize the absolute volume differences and express FA as a percentage of the average breast size [6]. This is known as relative fluctuating asymmetry: Proc. of SPIE Vol A-3

4 Relative Fluctuating Asymmetry: VOLUME relative FA = L "VOLUME R 2 # VOLUME L + VOLUME R ( ) (4) 2.3 Image Database 800 screening mammograms from the Digital Database of Screening Mammography (DDSM) were used for this study: 394 cancer cases, 324 benign cases, and 82 normal cases [23]. The normal cases were deemed normal based on 2-yr cancer-free follow-up. The benign and cancer cases were biopsy-proven and they contained all possible types of lesions (calcifications, masses, architectural distortions, asymmetries, etc.). Of those, 25 cancer, 74 benign, and 13 normal cases were excluded from further analysis because, upon visual inspection, the breast segmentation step was deemed unsuccessful in at least one view. A total of 688 mammograms were used to measure the FA. The average relative FA was compared among the 3 groups of cases (cancer, benign, normal) and across the 3 FA estimation methods (CC only, MLO only, CC+MLO). 3. RESULTS Table 1 reports the average breast volume fluctuating asymmetry for each pathology and estimation method. Table 1. Average breast volume fluctuating asymmetry for each group of cases (cancer, benign, normal) and for each estimation method implemented in this study. METHOD CANCER BENIGN NORMAL CC 0.045± ± ±0.020 MLO 0.047± ± ±0.019 CC+MLO 0.054± ± ±0.023 Overall, the results support that the average breast volume fluctuating asymmetry is significantly larger in patients with breast lesions than those without. However, fluctuating breast volume asymmetry was statistically significant between cancer and benign cases only based on MLO estimates. Table 2 summarizes the two-tailed p-values of the FA comparisons among the various groups of patients. In addition, fluctuating breast volume asymmetry measurements had substantially lower variability in normal patients than cancer and benign patients (see Table 1). In other words, even though some breast volume asymmetry is expected in women, the presence of a benign or cancerous growth appears to contribute further. 4. DISCUSSION Evolutionary studies have established asymmetry of paired structures as a strong indicator of developmental stability and good health. It is known that if an individual has harmful mutations, then the homeostatic mechanisms that maintain symmetry tend to break down. Our study was inspired by this basic principle of evolutionary and developmental biology. Since sexually selected traits (such as breasts) are more revealing of genetic or environmental stresses, we hypothesized that the presence of breast cancer may be manifested in higher breast asymmetry measured in a screening mammogram. Specifically, we hypothesized that mammographic analysis of bilateral breast asymmetry will provide us with a set of quantitative imaging biomarkers that correlate highly with the presence of breast cancer. The hypothesis does not contradict breast cancer etiology theories (genetic disease vs. the result of environmental causes). Proc. of SPIE Vol A-4

5 Preliminary analysis on a large set of screening mammograms supports that computer-based assessment of bilateral breast volume asymmetry could be exploited as an additional breast cancer risk biomarker during mammographic screening. Such biomarker can be used to alert accordingly the radiologist before mammographic interpretation. Further investigation with larger datasets is necessary to determine whether certain patient subgroups (e.g., unilateral/bilateral findings, masses vs. calcifications, etc.) present a stronger trend of bilateral breast size asymmetry as a diagnostic indicator of breast cancer. There were some important limitations with this pilot study. First, the quantitative biomarker depended on a single breast region segmentation algorithm. The study must be repeated with other breast segmentation techniques to determine the robustness of the discovered trends. Furthermore, the proposed measurements of fluctuating asymmetry could benefit from more sophisticated algorithms which could take into account the breast compression factor. Last but not least, the study could be repeated with other imaging modalities such as MRI for further validation of the underlying principle. Table 2. Two-tailed p-values of the differences in mean fluctuating breast volume asymmetry between groups of cases. * indicates statistical significance. METHOD GROUP 1 GROUP 2 2-tailed p-value CC cancer benign cancer normal 0.004* benign normal 0.014* MLO cancer benign 0.021* cancer normal <0.0001* benign normal <0.0001* CC+MLO cancer benign cancer normal <0.0001* benign normal <0.0001* REFERENCES [1] Thornhill, R. and Meller, A.P., "Developmental stability, disease and medicine," Biol Rev 72, (1997). [2] Manning, J.T. and Chamberlain, A.T., "Fluctuating asymmetry, sexual selection and canine teeth in primates," Proc Royal Soc London 251, (1993). [3] Meller, A.P. and Pomiankowski, A., "Fluctuating asymmetry and sexual selection," Genetica 89, (1993). [4] Meller, A.P., Soler, M. and Thornhill, R., "Breast asymmetry, sexual selection and human reproductive success," Ethol Sociobiol 16, (1995). [5] Scutt, D., Manning, J.T., Whitehouse, G.H., Leinster, S.J. and Massey, C.P., "The relationship between breast asymmetry, breast size and the occurrence of breast cancer," Br J Radiol 70, (1997). [6] Scutt, D., Lancaster, G.A. and Manning, J.T., Breast asymmetry and predisposition to breast cancer, Breast Cancer Res 8, R14 (2006). [7] Harvey, J.A., Fechner, R.E. and Moore, M.M., "Apparent ipsilateral decrease in breast size at mammography: A sign of infiltrating lobular carcinoma," Radiol 214, (2000). [8] Hilleren, D.J., Andersson, I.T., Lindholm, K. and Linnell, F.S., "Invasive lobular carcinoma - mammographic findings in a 10-year experience," Radiol 178, (1991). [9] Lau, T.K., and Bischof, W.F., "Automated detection of breast-tumors using the asymmetry approach," Comp Biom Res 24, (1991). [10] Yin, F.F., Giger, M.L., Doi, K.,, Vyborny, C.J. and Schmidt, R.A., "Computerized detection of masses in digital mammograms - automated alignment of breast images and its effect on bilateral-subtraction technique," Med Phys 21, (1994). Proc. of SPIE Vol A-5

6 [11] Mendez, A.J., Tahoces, P.G., Lado, M.J., Souto, M. and Vidal, J.J., " Computer-aided diagnosis: Automatic detection of malignant masses in digitized mammograms," Med Phys 25, (1998). [12] Zheng, B., Chang, Y.H., Gur, D., "Computerized detection of masses from digitized mammograms - comparison of single-image segmentation and bilateral-image subtraction," Acad Radiol 2, (1995). [13] Ferrari, R.J., Rangayyan, R.M., Desautels, J.E.L. and Frere, A.F., "Analysis of asymmetry in mammograms via directional filtering with Gabor wavelets," IEEE Trans Med Imag 20, (2001). [14] Rangayyan, R.M., Ferrari, R.J. and Frere, A.F., "Analysis of bilateral asymmetry in mammograms using directional, morphological, and density features, J Electron. Imaging. 16(1), (2007). [15] Bozek, J., "A Survey of Mammographic Image Processing Algorithms for Bilateral Asymmetry Detection", 51st International Symposium ELMAR-2009, September 2009, Zadar, Croatia, pp [16] Suri J. S. and Ragayyan R. M., Recent Advances in Breast Imaging, Mammography and Computer Aided Diagnosis of Breast Cancer. SPIE Press, April [17] Mirzaalian, H, Ahmadzadeh, M.R., Sadri, S. and Jafari, M., Pre-processing algorithms on digital mammograms, Proc. of the IAPR Conference on Machine Vision Applications (MVA), May 2007, Japan, pp [18] Katariya, R.N., Forest, A.P.M. and Gravelle, I.H., "Breast volumes in cancer of the breast," Br J Radiol 29, (1974). [19] Hoe, A.L., Mullee, M.A., Royle, G.T., Guyer, P.B. and Taylor, I., "breast size and prognosis in early breast-cancer," Ann R Coll Surg 75 (1), (1993). [20] Stomper, P.C., D'Souza, D.J., DiNitto, P.A. and Arredondo, M.A., "Analysis of parenchymal density on mammograms in 1353 women years old," AJR 167, (1996). [21] Fung, J.T., Chan, S.W., Chiu, A.N., Cheung, P.S. and Lam S.H., Mammographic determination of breast volume by elliptical cone estimation, World J Surg. 34, (2010). [22] Kalbhen, C.L., McGill, J.J., Fendley, P.M., Corrigan, K.W. and Angelats, J., "Mammographic determination of breast volume: Comparing different methods," AJR 173(6), (1999). [23] Heath, M., Bowyer, K., Kopans, D. et al, Current Status of the Digital Database for Screening Mammography, in Digital Mammography, Kluwer Academic Publishers (1998). Proc. of SPIE Vol A-6